Acoustic data-driven framework for structural defect reconstruction: A manifold learning perspective

Data-driven quantitative defect reconstruction using ultrasonic guided waves has recently demonstrated great potential in the area of non-destructive testing (NDT) and structural health monitoring (SHM). In this paper, a novel deep learning-based framework, called Deep-guide, has been proposed to convert the inverse guided wave scattering problem into a data-driven manifold learning progress for defect reconstruction. The architecture of Deep-guide network consists of the efficient encoder-projection-decoder blocks to automatically realize the end-to-end mapping of noisy guided wave reflection coefficients in the wavenumber domain to defect profiles in the spatial domain by the manifold distribution principle and intelligent learning. Towards this, results by the modified boundary element method for efficient calculations of scattering fields of guided waves have been generated as acoustic emission signals of the Deep-guide to facilitate the training and extract the features homeomorphically. The correctness, robustness and efficiency of the proposed framework have been demonstrated throughout several examples and experimental tests of circular defects. It has been noted that Deep-guide has the ability to achieve the high-quality defect reconstructions and provides valuable insights into the development of effective data-driven techniques for structural health monitoring and complex defect reconstructions.